COMMITTEE CHAIR: Dr. Ahmed Ahmed
CO-COMMITTEE CHAIR: Dr. Md. Jobair Bin Alam

TITLE: MULTI-TASK GROWING INTERPRETABLE NEURAL NETWORK FOR MULTI-TARGET SYMBOLIC REGRESSION

ABSTRACT: This research presents a comprehensive, three-phase hierarchical data collection framework integrating Internet of Things (IoT) sensors, aerial RGB imagery, and Electrical Resistivity Imaging (ERI) to correlate multi-scale hydro-geophysical soil parameters for comprehensive slope stability assessment. The first phase established a laboratory-scale physical slope model using Fat Clay (CH) to replicate failure-prone conditions. The prototype was instrumented with distributed IoT sensors measuring tilt, volumetric moisture content, and soil matric suction. Real-time data was transmitted to a custom cloud-based web graphical user interface (GUI) for remote visualization. Controlled artificial rainfall simulations validated the system’s sensitivity to subtle kinematic and hydrological changes, establishing a foundational point-scale measurement infrastructure. To transition from lab-scale point measurements to vast field-scale analysis, the second phase focused on surface data extraction. Close-range aerial RGB imagery was captured for a targeted field location to extract multiple quantitative optical indices (e.g., Green Leaf Index (GLI), Visible Atmospherically Resistant Index (VARI), Normalized Green-Red Difference Index (NGRDI), etc.). The index exhibiting the highest statistical correlation score was selected to map surface-level variations. This optical data collection validated the visual monitoring infrastructure and serves as a foundational proxy for future, large-scale Unmanned Aerial Vehicle (UAV) deployment. The third phase conducted ERI surveys at the same field location to capture subsurface resistivity profiles. The core of this research establishes a direct data correlation framework between the extracted surface features (RGB) and subsurface geophysical conditions (soil resistivity). Utilizing Pearson correlation to assess linear relationships and Spearman rank correlation to evaluate monotonic trends, these distinct spatial datasets are systematically aligned. Quantifying these statistical relationships allows for the identification of underlying soil behavior patterns, enabling an in-depth analysis of near-surface anomalies and potential landslide triggers. This multi-scale approach bridges the critical gap between controlled laboratory validation and operational field mapping. By hierarchically correlating continuous IoT point-sensing with expansive RGB and ERI spatial data, this research delivers a robust, scalable data collection infrastructure. This framework addresses the limitations of conventional techniques, providing a cost-effective, multi-tiered solution for proactive anomaly detection and landslide risk mitigation.

Keywords: AI, IoT, UAV, ERI, Data Fusion

Room Location: S. R. Collins. Room 111: CS Main Conference Room.